Orion Trip Generator

• takes in JSON S3 files (API output) to generate trips

What is a trip?

• uniqueness is defined by (vehicle_id, route_id, direction_id) staying consistent with every transit state.
• If this tuple is different in the next state, then we end the trip. If we see a tuple that we didn't see last state, we start a new trip.
• when a trip ends, we write it to a JSON file in S3 (agency_id-vehicle-id_route-id_direction_id-start_time-end_time.json, where times are in a timestamp in milliseconds). We can worry about compressing it or loading it into some databases later.

What is a state?

• it's a snapshot of where all Muni vehicles are with a timestamp

Why?

• our goal is to get the speed & reliability of routes.

• An easy way of approaching this problem is to look at individual trips and then get data about them along route segments.

• Storing trip metrics also makes it easier to eventually quickly handle more complex analysis due to aggregated metrics.

• our goal is to make Muni's GPS data open and accessible to use.

• making trip data accessible via S3 keeps individual file sizes small and is straightforward & logical for open-data users to retrieve.

What the output will look like:

agency:
startTime:
endTime:
route:
direction:
vid:
states: [{
    vtime:
    lat:
    lon
}]

(json)

How it'll work:

• how it'll persist state:

• each time it goes to the next state, it dumps its state

Design

• Read raw s3 vehicle data from s3 on first startup and create trips (unique tuple in memory)
• Write trips to state file on disk (or s3?) on first startup
• When new raw vehicle file is put in s3, publish sqs message which orion-trip-generator is listening for
• orion-trip-generator consumes sqs message which triggers: -- updating trips in memory (states array) -- write trips that no longer exist to agency trip s3 bucket and remove them from memory -- add new trips to memory. -- write current trips state to state file -- all of above should be atomic
• if orion-trip-generator restarts, it gets it's state from state file
• if state file doesn't exist it read latest raw file to get state

Possible Issues

• If there is a lag between publishing raw files (hours?, days?) this could lead to trips that appears to have taken too long to complete.
• If the trip generator crashes and is restarted after hours or days, the same issue as above could occur (trips that appear to take days to complete)
• Should we ignore writing trips that last longer than a certain period of time?
• Should we still write suspect trips but mark them as possibly invalid?